nomenklatura
Nomenklatura de-duplicates and integrates different Follow the Money entities. It serves to clean up messy data and to find links between different datasets.
Usage
You can install nomenklatura via PyPI:
$ pip install nomenklatura
Command-line usage
Much of the functionality of nomenklatura can be used as a command-line tool. In the following example, we'll assume that you have a file containing Follow the Money entities in your local directory, named entities.ijson. If you just want try it out, you can use the file tests/fixtures/donations.ijson in this repository for testing (it contains German campaign finance data).
With the file in place, you will cross-reference the entities to generate de-duplication candidates, then run the interactive de-duplication UI in your console, and eventually apply the judgements to generate a new file with merged entities:
# generate merge candidates using an in-memory index:
$ nomenklatura xref entities.ijson
# note there is now a sqlite database, `nomenklatura.db` that contains de-duplication info.
$ nomenklatura dedupe entities.ijson
# will pop up a user interface.
$ nomenklatura apply entities.ijson -o merged.ijson
# de-duplicated data goes into `merged.ijson`:
$ cat entities.ijson | wc -l
474
$ cat merged.ijson | wc -l
468
The resolver graph database location can be customised by setting the environment variable NOMENKLATURA_DB_URL
Programmatic usage
The command-line use of nomenklatura is targeted at small datasets which need to be de-duplicated. For more involved scenarios, the package also offers a Python API which can be used to control the semantics of de-duplication.
nomenklatura.Dataset - implements a basic dataset for describing a set of entities.nomenklatura.Store - a general purpose access mechanism for entities. By default, a store is used to access entity data stored in files as an in-memory cache, but the store can be subclassed to work with entities from a database system.nomenklatura.blocker.Index - a cross-reference blocker for correlating entities inside of a dataset, or across different datasets.nomenklatura.Resolver - the core of the de-duplication process, the resolver is essentially a graph with edges made out of entity judgements. The resolver can be used to store judgements or get the canonical ID for a given entity.
All of the API classes have extensive type annotations, which should make their integration in any modern Python API simpler.
Design
This package offers an implementation of a data deduplication framework centered around the FtM data model. The idea is the following workflow:
- Accept FtM-shaped entities from a given source (e.g. a JSON file, or a database)
- Build an inverted index of the entities for dedupe blocking
- Generate merge candidates using the blocking index and FtM compare
- Provide a SQL persistence abstraction for merge challenges and decisions
- Provide a text-based user interface to let users make merge decisions
- Export consolidated entities that cluster source entity data
The Enrichment framework enables linking entities to records in other data sources, and enriching them with information from those records.
Later on, the following might be added:
- A web application to let users make merge decisions on the web
Resolver graph
The key implementation detail of nomenklatura is the Resolver, a graph structure that manages user decisions regarding entity identity. Edges are Judgements of whether two entity IDs are the same, not the same, or undecided. The resolver implements an algorithm for computing connected components, which can the be used to find the best available ID for a cluster of entities. It can also be used to evaluate transitive judgements, e.g. if A <> B, and B = C, then we don't need to ask if A = C.
Reading
Contact, contributions etc.
This codebase is licensed under the terms of an MIT license (see LICENSE).
We're keen for any contributions, bug fixes and feature suggestions, please use the GitHub issue tracker for this repository.
Nomenklatura is currently developed thanks to a Prototypefund grant for OpenSanctions. Previous iterations of the package were developed with support from Knight-Mozilla OpenNews and the Open Knowledge Foundation Labs.
Releasing
bump-my-version bump patch
git push && git push --tags
